"I stumbled on a giant cow artificial intelligence course, could not help but share to everyone. Tutorial is not only a zero-based, user-friendly, and very humorous, like watching a fiction! Think too much bad, so to others. Click here to jump to the tutorial. "
At present the main trends in the study based on the recommendation system from a single, independent recommendation system algorithm gradually integrated hybrid recommendation algorithm direction to a combination of a variety of recommendation algorithm development, combined with a growing number of users tag data, social network data, context information, location information. Recommended group has become a very hot topic at present. And some also used outside of the recommendation algorithm system in the field, such as the concept of fuzzy sets, genetic algorithms in the field of artificial intelligence, Bayesian networks. Of course, there are some studies are constantly dig traditional, classic, such as content-based filtering, collaborative filtering algorithms may be improved algorithms.
Many data distribution on the Internet are all long-tailed distribution, frequency of use most of the data is very low, only a small part of the data is widely used, this small part of the data is actually popular data, but if the data using the popular and unpopular data amount for comparison, you will find popular usage data is far greater than the amount of hot data, and perhaps people think the opposite, but this is after a long period of statistical analysis concluded. User behavior data is actually in line with the law, regardless of the popularity of the article from the article's point of view, or user popularity of the user's point of view are in line with long-tailed distribution.
Overview recommendation system algorithm
Research recommendation system can be divided into three phases, the first is based on the traditional service, the second stage is based on the current social networking service, and the third stage is the upcoming things. Which it generated a lot of basic and important algorithms, such as collaborative filtering (including those based on and on items the user), based on the recommendation algorithm, the hybrid recommendation algorithm content-based recommendation algorithm statistical theory based social network information (attention , attention, trust, visibility, credibility, etc.) of the filtering algorithm, community recommendation algorithm, location-based recommendation algorithm. Which collaborative filtering recommendation algorithm based on neighborhood is the recommended system is the most basic, most central, most important algorithm not only get more in-depth research in academia, but also in the industry have also been very widely used, based on the neighborhood algorithm is divided into two categories, one is user-based collaborative filtering algorithm, the other is collaborative filtering algorithm based on item, in addition, based on the recommendation algorithm goods is also very wide, so it will be below three kinds of basic algorithm is described in detail.
Based on user collaborative filtering algorithm
Based on collaborative filtering algorithm referred to as users, and its simple application scenario is: when the user needs personalized recommendations, you can find with him is similar to other users (through interests, hobbies or behavioral habits, then those users like and they do not know items recommended to the user.
Note: The two users with similar interests on popular items, does not mean they have similar interests, this time to increase the punishment.
Collaborative filtering algorithm based on an article
Based on collaborative filtering algorithm for short items, its simple application scenario is: when a user needs personalized recommendation, for example because before he bought Jin Yong's "Legend of the Condor Heroes" book, so he would recommend "evil "because a lot of other users have also purchased the two books.
Note: 1, if it is popular items, many people like it, it will be close to 1, it will cause a lot of popular items and items are similar, this time to increase the punishment.
2, the contribution of active users is less than the similarity of goods inactive users.
Content-based recommendation algorithm
though is more popular collaborative filtering recommendation algorithm, has extensive research and use in academia and industry, but the basis of the same algorithm as the field one by one based on the recommendation system for recommending content is also very important, other recommended algorithms it is the earliest. The basic principle is based on the historical behavior of the items before the user (such as user purchased what items, what collection of objects too, has rated and so on, then according to the calculation of these items similar articles, and recommend them to the user. For example Jin Yong's novels bought before the user, which can explain the user may be a Jinyong Mi or Wuxia Mi, then we can recommend some other martial arts novels of Jin Yong to the user. before recommendation algorithm based on content has become a content-based filtering (search ) algorithm, used in the early days of information retrieval and information through
Content-based recommendation algorithm generally includes the following three steps:
1, some of the features extracted for each item to indicate that the article.
2, using the user's historical behavior data analysis features of these items, and to learn the characteristics of the user's preferences or interests.
3, by comparing the user interests and characteristics to be obtained in step recommended items, determining a set of maximum correlation as a recommendation list items.
Contrast algorithm based recommendation system
UserCF is for users to recommend which items the user and his mutual interest like, ItemCF is recommended for users similar to those of his previous favorite items items, so UserCF recommend a more socialized, that is, the recommended items are consistent with the user's interest that group of popular items, ItemCF recommended more personality, because most of the items are recommended to meet their own unique interests.
----------------
Disclaimer: This article is the original article CSDN bloggers "Hehe Ming", and follow CC 4.0 BY-SA copyright agreement, reproduced, please attach the original source and link this statement.
Original link: https: //blog.csdn.net/u011630575/article/details/80171597
